基于粒子群算法的原位体绘制参数设置算法

Auto Tunning for In-Situ Volume Depth Image Generation Based on Particle Swarm Optimization Algorithm

原位可视化是解决千万亿次科学计算数据分析的最有效途径。在原位进行体绘制时,使用体深度图像作为中间表示是一种备受关注的方法,但该方法的参数选择较为困难。对体深度图像的生成时间、数据压缩率和绘制质量三个指标进行了全面分析,确定了各参数对这些指标的影响方式,给出了一套可调控的评估体系,并利用粒子群算法的快速收敛性质在参数空间进行参数寻优,来自动设置绘制参数组。实验结果表明,该方法可以自动获取到最优参数组,而且速度比简单的网格搜索方法快一个数量级以上。

In-situ visualization is the most effective way for data analysis of peta-scale scientific computation. Volume Depth Image(VDI)is a promising method for in-situ volume rendering, while its parameter selection is a crucial nodus.After analysis of three main targets of VDI:generation time, data compression rate and rendering quality, this paper presents an auto parameter selection approach based on Particle Swarm Optimization(PSO)algorithm. First a set of evaluation functions is designed based on analysis of the effect of parameters to targets. Then the fast convergence property of the particle swarm algorithm is utilized to tune the parameters automatically for the generation of VDI. Experimental results show that this approach can get the optimal parameters set automatically and it is at least one magnitude faster than the grid search method.